Abstract: When Two Wrongs Don’t Make a Right: Examining Confirmation Bias and the Role of Time Pressure During Human-AI Collaboration in Computational Pathology

Artificial intelligence (AI)-based decision support systems hold the potential to enhance diagnostic accuracy and boost efficiency in computational pathology. However, human-AI collaboration can also give rise to new cognitive biases or intensify existing ones, such as confirmation bias driven by false confirmation, where erroneous human judgments are reinforced by inaccurate AI outputs. This effect may be exacerbated under time pressure, a pervasive factor in routine pathology that places strain on medical experts’ cognitive resources. In this study [1], we quantified confirmation bias arising from false confirmation through AI suggestions and examined the moderating influence of time constraints in a web-based experiment involving pathology experts (n = 28) estimating tumor cell percentages. Our results indicate that AI integration fuels confirmation bias, as evidenced by a statistically significant positive linear mixed-effects model coefficient showing that greater alignment between AI recommendations and erroneous human estimates subsequently increases agreement with system advice. Conversely, time pressure appeared to mitigate this dynamic. These findings highlight potential pitfalls of medical AI and contribute to establishing a foundation for the safe implementation of AI-based decision support systems into clinical practice.

Technische Hochschule Ingolstadt
Flensburg University of Applied Sciences